Add integration tests that run against a real Redshift cluster

[JoshRosen]

This PR adds end-to-end integration tests that run against a real Redshift cluster.

I moved the build definition from .sbt to .scala format in order to use SBT's standard IntegrationTests configuration.

I configured the Travis build to store Redshift credentials using encrypted environment variables. To minimize security risks, Travis does not allow pull requests from third-party forks to run tests using these credentials.

[codecov-io]

Current coverage is 85.90%

Merging #41 into master will not affect coverage as of 6dff0a2

@@            master    #41   diff @@
=====================================
  Files            9      9       
  Stmts          305    305       
  Branches        70     70       
  Methods          0      0       
=====================================
  Hit            262    262       
  Partial          0      0       
  Missed          43     43       

Review entire Coverage Diff as of 6dff0a2

Powered by Codecov. Updated on successful CI builds.

[JoshRosen]

Alright, the basic test is now passing (fixed the non-deterministic failure issue), so I'm now going to focus on enabling more tests.

[JoshRosen]

I wonder if I should purposely give these different names than the ones expected by the Amazon SDKs; this might be necessary in order to be able to test credential mechanisms (e.g. for writing a regression test for #32)

[JoshRosen]

All tests are now enabled, but some seem to fail due to problems in the tests themselves (due to not using a HiveContext) or due to bugs.

[JoshRosen]

This is now ready for review. Let's not commit until #42 goes in and until I have time for one final pass. This isn't all of the tests that I want to add but it's a reasonable first cut; I'll add more tests and fix more potential bugs in smaller followups.

[marmbrus]

Nit: is there a reason to not test style and other fast stuff first?

[JoshRosen]

I wanted the tests to still be able to run even if style checks failed. I figured this wasn't a huge deal here compared to what we do in Spark because the test time is really fast.

[marmbrus]

nit: newline

[marmbrus]

This is awesome! LGTM.

[JoshRosen]

This was rebased on top of #32, but I'm going to try to back out of that rebase so that we can write a proper regression test for that issue.

[JoshRosen]

Alright, I have now backed out all non-test-related changes and have addressed the review comments, so I'm going to merge this now and will continue to iterate in followups.